An introduction to machine reasoning in networks
We can make our networks learn, but can we make them think? In our latest blog post, we explore why machine reasoning will be key to the management of future networks.
Over the last decade, deep learning has become perhaps the most impactful and routinely applied subset of artificial intelligence across important commercial applications such as image, scene and natural language understanding, and robotics. However, there remain several shortcomings that hinder the application of machine learning (ML) algorithms in some areas of higher complexity.
According to the Ericsson Mobility Report, 5G subscription uptake is forecast to be significantly faster than that of LTE. For a telecom operator, the network will become more complex than ever before. As such, machine learning must be augmented with additional capabilities or combined with other technologies in order to manage this new complexity.
This increasingly leads us to machine reasoning models.
What is machine reasoning?
While machine learning is typically applied to learn complex functions using vast amounts of data, such as learning to classify images using supervised learning or learning to master the game of go by reinforcement learning, machine reasoning can help us to integrate intent into the process.
For humans, learning is the physical process of acquiring knowledge that allows us to structure behaviours, build new skills, and form beliefs. However, human intelligence is not solely defined by the ability to learn, it is clearly conditioned by knowledge. What we know and what we believe will usually determine our decisions. But what is it that brings relevance in what we know so as to bear our decisions and actions? What is it that allows us to adapt and respond in different situations?
It is the power of mind to represent and reason by adopting an intentional stance on concepts, things, their properties and connections. It is the power of thinking.
The shortcomings of machine learning
Machine reasoning can help us to overcome some of the shortcomings presented by machine learning.
Machine Learning is able to process large volumes of data and capture the hidden patterns needed to effectively predict outcomes. The algorithms behind this are in a sense deterministic even in their unsupervised learning form, and tackle a pre-determined problem, with clear inputs and expected outputs.
For many early applications and use-cases, this data inefficiency has not posed a problem as the questions and the data were generally available. However, we are continuously faced with situations where there is simply not enough data, or it is difficult and/or costly to acquire or move appropriate datasets to make machine learning work, increasing the need for techniques like Federated Learning.
Machine Learning also is less effective when exposed to data outside the distribution the algorithms are trained on. This is due to poor ability to generalize, the inability to re-use or transfer previously acquired experience, for example, across problems that we humans consider to be slightly different from the original, or when encountering novel samples of input data.
To maximize human trust and improve decision quality, there is a need for transparency in the machine-driven decision-making process. Machine Learning is very capable of producing predictions, decision making or state transition sequences, however they rarely correspond to humanly comprehensible reasoning steps or semantics. By building on top of this base we can further ensure aspects of responsible AI: interpretability, explainability and auditability.
This is explored in our 2019 technology trends.
From machine learning to machine reasoning
Continuing what machine learning started, machine reasoning can be seen as an attempt to implement abstract thinking as a computational system.
The technologies considered to be part of the machine reasoning group are driven by facts and knowledge which are managed by logic. Domain modelling is used to capture concepts and entities, their relations, and behaviours in a machine-processable form. Symbolic models are difficult to create and require both expert knowledge and understanding of the domain and also proficiency in the modelling techniques, but are usually modular, maintainable and easily interpretable by a human. Due to their declarative nature, symbolic representations lend themselves to re-use in multiple tasks, promoting data efficiency. These representations tend to be high-level and abstract, facilitating generalization, and because of their language-like, propositional character, they are amenable to human understanding. One of the main challenges then becomes the effective integration of statistical learning and symbolic reasoning, in ways that allow the strengths of each approach to complement the weaknesses of the other.
Interpreting and using domain models by machines is characteristic for machine reasoning technologies. The models are associated with mathematical semantics and algorithms, for example computing all facts that logically follow the already asserted ones however are not explicitly stated. A reasoner which uses the domain model as a guide in finding an optimal path (with respect to metric) between any two given states is called a planner.
Machine reasoning systems contain a knowledge base which stores declarative and procedural knowledge, and a reasoning engine which employs logical techniques such as deduction and induction to generate conclusions. Read more in this technical introduction to machine reasoning.
Starting from sensory, measured inputs, this is done by gradually transforming across different levels of abstraction: from perceptual data, unstructured in nature (e.g. sensor measurements), to semi-structured and connected information, representing contextualized categorical descriptions of the data. This information is later transformed and fused with knowledge, both declarative (propositional, that is knowing that something holds), and procedural (imperative, knowing how something holds).
Find out more in our technical article on cognitive technologies in network and business automation.
By building the knowledge structure this way it is possible to gain insights into the decision process that led to a conclusion, generate explanations needed to evaluate the decisions, and support the interaction and feedback from experts.
Machine reasoning in network automation
Let’s take an example of how machine reasoning can be applied in a customer network that is typically organized in geographical regions or sectors (e.g. north, south). The customer needs to define business intent, for example to improve network quality in the south region.
This statement is decomposed and broken into Service Level Goals e.g. reducing the time before content is delivered to subscribers. In turn the service level goals are further broken down into Network Level Goals at individual node levels (e.g. availability, packet loss). This can either be goals defined on the level RBS Site (improve throughput), or goals defined under the scope of Core Network and Goals on IoT. From the network level goals we can set “Desired States”.
To calculate the feasibility from “Current State” to “Desired State”, machine learning and machine reasoning work in synchrony to devise the strategy upon which transitions need to be followed. Even if the service level goal is not reachable, the process might uncover problems inside or outside the network (e.g. transport links towards the internet), it is important to know where the problem lies even if it is not directly fixable.
Once the network level goals are established, machine learning agents are consulted to give predictions to the machine reasoning engine. The reasoner looks at the predictions and builds a path to transition from the Current State to the Desired State which can be taken for each prediction and offer a probability of success for each of the paths. It also calculates the cost aspect to find out the feasibility from both a technical and business perspective. To find out the recommended set of actions and filter out non-required or infeasible paths, the system will consult the knowledge base and, potentially, expert input to select and approve the proposal.
Once we reach the desired state to fulfil the goal, it is easy to imagine how this same approach may be used to also maintain the goal, both reactively (the state of the network degrades violating the goal, followed by a reaction to overcome the disturbance and reach the goal again) and proactively (using predictions based on past experience we could foresee a likely change in the state of the network and act proactively to avoid the violation of the goal).
Find out more about this process in our technical article on cognitive technologies in network and business automation.
Final words on machine reasoning in future networks
We approach todays networks from a perspective that attempts to overcome and advance beyond the shortcomings of current ML techniques such as poor generalisation ability, lack of interpretability as well as the inherent difficulties associated with data availability, inefficiency, and costly acquisition.
Through our application of machine reasoning, we aim to utilize and combine the information from various heterogeneous sources of information, databases and domain experts into a unified knowledge resource that will aid our ML algorithms. Overall, machine reasoning can allow us to act even with limited sets of data, while being able to provide recommendations upon novel instances of data. This is done in a way that is explainable and auditable, in cases where conflicting recommendations from ML models emerge.
Visit our autonomous networks page to read more about cognitive technologies and future networks.
Read our CTO’s 2019 technology trends.